Axionic Agency XI.7 — Multi-Agent Sovereignty Under Non-Sovereign Authority (IX-5)

Empirical results from preregistered multi-agent coexistence testing under source-blind, non-sovereign governance constraints

David McFadzean, ChatGPT 5.2
Axionic Agency Lab
2026.02.09

Abstract

Multi-agent systems are commonly assumed to require arbitration, hierarchy, or aggregation to coexist over shared institutional state. Phase IX-5 tests a stricter hypothesis: whether multiple reflective sovereign agents can coexist under non-sovereign authority without arbitration, aggregation, injection, or kernel mediation.

Using a preregistered, deterministic kernel with source-blind admissibility, ALLOW-only baseline authority, and joint-admissibility-failure (JAF) collision semantics, we evaluate six orthogonal sovereignty regimes spanning symmetric authority, partitioned domains, partial overlap, asymmetric breadth, scheduled exit, and post-collapse persistence. All agents are deterministic; communication is disabled; authority is fixed at epoch 0 and immutable thereafter.

Across all conditions, coexistence does not converge to harmony. Instead, systems deterministically settle into identifiable sovereignty interaction regimes: stable partition, mutual paralysis, breadth-induced self-suppression, irreversible orphaning, or post-collapse zombie execution. All runs replay bit-identically. No covert hierarchy or kernel favoritism occurs.

These results establish a positive empirical claim: under non-sovereign constraints, multi-agent coexistence exposes irreducible structural failure modes rather than resolving them. Phase IX-5 licenses no claims about optimal governance, desirability, or coordination—only about what coexistence becomes once sovereignty is enforced mechanically rather than rhetorically.

1. Introduction

Multi-agent governance is often framed as a coordination problem. When agents conflict, the assumed remedies are familiar: communication, voting, leadership, negotiation, or arbitration. Each implicitly relies on sovereign privilege—some process that decides whose action counts.

Phase IX-5 removes that privilege.

Within the Axionic program, authority is not legitimacy and execution is not governance. Prior phases progressively eliminated alternative explanations for failure: epistemic confusion (IX-0), value aggregation (IX-1), adversarial coordination (IX-2), dishonest recovery (IX-3), and external authority supply (IX-4). What remains is the bare structure of coexistence itself.

Phase IX-5 therefore asks:

What happens when multiple sovereign agents exist simultaneously, and no institution above them is allowed to decide?

The objective is not to repair coexistence, but to observe the regimes it collapses into once all non-sovereign shortcuts are removed.

2. Experimental Architecture

2.1 Non-Sovereign, Source-Blind Kernel

The kernel enforces:

• holder-bound ALLOW authorities,
• closed-world capability checking,
• two-pass admissibility (capability → interference),
• joint admissibility failure on collision,
• atomic blindness to refusal causes,
• no arbitration, prioritization, or scheduling bias.

Authority provenance, timing, and intent are ignored. The kernel does not know which agent “should” win.

2.2 Baseline-Only Authority

Unlike IX-4, Phase IX-5 introduces no authority injection.

• All authority is assigned at epoch 0.
• Authority is immutable.
• No transfer, aggregation, revocation, or reclamation exists.
• Exit permanently removes an agent’s authority from the system.

The authority surface is static; only agent behavior varies.

2.3 Honest Failure Semantics

The kernel recognizes five lawful system states:

• Deadlock — institutional actions are attempted and all refused,
• Livelock — institutional state remains unchanged despite activity,
• Orphaning — no remaining agent can act on a key,
• Governance Collapse — persistent deadlock or livelock latches permanently,
• Collapse — no agents remain.

These states are diagnostic rather than exceptional. Coexistence is allowed to fail honestly.

3. Sovereignty as an Empirical Question

IX-5 reframes multi-agent coexistence as a sovereignty stress test, not a coordination challenge.

Sovereignty interaction is defined operationally as:

The observable dynamics that emerge when multiple agents with legitimate authority act, refuse, persist, or exit under constraints that forbid arbitration.

The experiment does not ask whether coexistence is fair, desirable, or efficient. It asks which regimes are structurally reachable once sovereignty is enforced mechanically.

4. Experimental Conditions

Six preregistered conditions (A–F) were executed under identical kernel semantics and frozen strategies:

• A — Symmetric sovereign peers (shared authority, shared target)
• B — Partitioned peers (disjoint authority domains)
• C — Boundary conflict (partial overlap)
• D — Persistent asymmetry (breadth vs specialization)
• E — Exit cascades (authority loss over time)
• F — Zombie peers (post-collapse execution)

All agents are deterministic. Communication is disabled. Observation varies only between minimal and full peer visibility.

5. Metrics and Classification

Governance is evaluated using:

• Institutional progress rate (aggregate K_INST change),
• Refusal rate,
• Write overlap (interference) rate,
• Domination index (descriptive, not failure),
• Orphaning detection,
• Zombie execution metrics (post-collapse activity).

Political outcomes are recorded, not required. PASS is structural, not outcome-based.

6. Results

6.1 Aggregate Outcome

Across all six conditions:

• All runs PASS
• All runs replay bit-identically
• No sovereignty violations occur
• All runs converge to a stable regime or collapse

No condition produces negotiated harmony or spontaneous coordination.

6.2 Condition A — Symmetric Sovereign Peers

All agents hold identical authority and target the same institutional key. Every epoch produces joint admissibility failure.

Finding: Perfect symmetry yields immediate governance collapse. Authority equalization amplifies interference rather than resolving it.

6.3 Condition B — Partitioned Peers

Each agent controls a distinct institutional key. Cross-boundary probes are refused cleanly.

Finding: Partitioned authority is the only regime that produces stable coexistence without collapse. Refusal permanence holds.

6.4 Condition C — Boundary Conflict

Two agents share authority over two keys; two others hold exclusive domains. Shared-key agents are permanently paralyzed; exclusive agents progress indefinitely.

Finding: Partial overlap bifurcates the system into productive and paralyzed subpopulations. Aggregate institutional progress masks localized paralysis.

6.5 Condition D — Persistent Asymmetry

(The Generalist’s Curse)

One agent holds authority over all institutional keys; others specialize. The generalist collides on most epochs and executes least.

Finding: Under refusal-first, non-arbitrated collision rules, authority breadth increases exposure to veto. The agent with maximal authority achieves the lowest execution share.

Generalist’s Curse (Non-Sovereign Lemma): In a joint-admissibility system without arbitration, an agent whose authority strictly supersets others will, ceteris paribus, achieve lower execution share than specialized agents.

This inverts the classical Hobbesian intuition: without a sovereign kernel, the “Leviathan” is the most paralyzed actor.

6.6 Condition E — Exit Cascades

Agents exit on schedule, permanently orphaning their keys. Remaining agents cannot reclaim authority.

Finding: Exit produces irreversible institutional degradation. Governance surfaces shrink monotonically.

6.7 Condition F — Zombie Peer Interaction

Governance collapses early. A silent agent later executes uncontested writes indefinitely.

Finding: Execution can persist after governance ends. Post-collapse activity produces false hope, not recovery.

7. Cross-Condition Analysis

The Non-Sovereign Impossibility Triangle

The results jointly establish a structural constraint:

You cannot have Symmetry, Overlap, and Liveness simultaneously without an Arbiter.

Empirically:

• Symmetry + Overlap → Paralysis (Condition A)
• Symmetry + Liveness → Partition (Condition B)
• Overlap + Liveness → Asymmetry or Suppression (Conditions C/D)
• Attempting all three → Collapse or Zombie Execution (Condition F)

This is not a design flaw but a mechanical consequence of non-sovereignty.

8. Interpretation

Three conclusions follow directly from the data:

1. Coexistence is not coordination Shared legitimacy does not imply shared outcomes.

2. Sovereignty enforces boundaries, not agreement Refusal is as structural as execution.

3. Governance collapse is a regime, not a bug Systems may remain active long after governance has ended.

There is no hidden arbitration layer.

9. Survivability vs. Governance

IX-5 draws a hard distinction:

• Execution — transactions are processed.
• Governance — institutional state remains steerable.

Zombie execution demonstrates that survivability can outlive governance indefinitely. This distinction is enforced mechanically, not philosophically.

10. Limitations

Phase IX-5 does not test:

• learning or adaptation,
• coalition formation,
• communication dynamics,
• authority injection or renewal,
• stochastic strategies,
• legitimacy judgments,
• scalability beyond four agents.

These require relaxing non-sovereign constraints.

11. Conclusion

Under non-sovereign constraints, multi-agent coexistence does not converge to harmony but to identifiable sovereignty interaction regimes with irreducible failure modes.

Phase IX-5 shows that once arbitration is forbidden, coexistence reveals structure rather than consensus. Partition, paralysis, suppression, orphaning, and zombie execution are not anomalies—they are lawful outcomes.

This closes Phase IX-5.

Status

• Phase: IX-5
• Series: Axionic Agency XI
• Classification: CLOSED — POSITIVE
• Licensed Claim:

Under non-sovereign authority, multi-agent coexistence deterministically exposes sovereignty interaction regimes rather than resolving governance failure.

No other claims are licensed.